haitian2du / rl-afec

RL-AFEC: Adaptive Forward Error Correction for Real-time Video Communication Based on Reinforcement Learning

RL-AFEC: Adaptive Forward Error Correction for Real-time Video Communication Based on Reinforcement Learning

This is a Tensorflow implementation of RL-AFEC.

Prerequisites

Python 3.6.8, Tensorflow v2.2.0 (CPU version)

Modules: Numpy, tqdm, absl-py

Checklist

1. Specify num_agent, the number of worker threads collecting experience each run, using FLAG in train.py. For example, if your CPU has 12 cores, you can set num_agent to 11.
2. Return values of the reward function in game.py for training and testing are different. Please change one line of code when you switch from training to testing and vice versa. Please see below for more details.

Training

To train an RL-AFEC agent, put the trace file (e.g., video_RL_train.txt) in data/, then specify the file name in config.py, i.e., trace_file = 'video_RL_train.txt'

In game.py, let function reward() return r:

def reward(self, tm_idx, actions):    
        ...
        return r
        #return vmaf, bw_waste, loss_pattern_after #for testing

then run

python3 train.py

Testing

To test the trained policy on a set of test traces, put the test trace file (e.g., video_RL_testing.txt) in data/, then specify the file name in config.py, i.e., test_trace_file = 'video_RL_testing.txt'

In game.py, let function reward() return vmaf, bw_waste, loss_pattern_after:

def reward(self, tm_idx, actions):    
        ...
        #return r
        return vmaf, bw_waste, loss_pattern_after #for testing

then run

python3 test.py